PHD DEFENSE Tuesday, September 30th, 2003, 2:00 pm, Auditorium, IRIT

A diagnosis approach based on Fuzzy Logic Application to car engine dyno test benches (Une approche du diagnostic basée sur la logique floue – Application aux bancs d'essais de moteurs automobiles.)

by Olivier de Mouzon in front of the following jury: Dr Dr Pr Dr Pr Mr Dr Dr

Bernadette Serge Claudette Didier Laurent Christophe Henri Louise

Bouchon-Meunier Boverie Cayrol Dubois Foulloy Peyrau Prade Travé-Massuyès

CNRS Research Advisor, LIP6, Paris (France) Research and Development manager, Siemens VDO Automotive Paul Sabatier University (Toulouse III, France), IRIT CNRS Research Advisor, IRIT, Toulouse (France) Savoie University, Annecy (France) Siemens VDO Automotive engine dyno test bench manager CNRS Research Advisor, IRIT, Toulouse (France) CNRS Research Advisor, LAAS, Toulouse (France)

Referee Member President Guest Referee Guest Supervisor Member

Key-words : fault diagnosis, uncertainty, fuzzy logic, possibility theory, abduction, experts’ knowledge, multiple faults, engine dynamometer. Summary : This study presents a fuzzy logic framework for fault diagnosis problems. It has been applied in a real-world problem: Siemens VDO Automotive engine dyno tests benches, Toulouse (France). The diagnosis is based on observations and pieces of knowledge. In this work, observations may be imprecise or even uncertain. As for the knowledge, it is based on human beings (engine dyno test benches experts), whose uncertainty has to be taken into account. For this reason, the study is based on possibility distributions expressing the effects of faults on measured (or calculated) attributes or the normal behavior of these attributes. This approach also is a bridge between Artificial Intelligence and Automatic communities. Basically, the diagnosis makes use of two indices: • One for consistency (deduction) ; • The other for relevance (abduction). Finally, some extensions are presented: • Twofold fuzzy sets for a better knowledge formalization when dealing with uncertainty ; • Multiple fault diagnosis ; • Cascading faults (by means of a cascade graph or, for more temporal aspects, by means of chronicles). Some (extension less) prototypes have been implemented, from the (off-line) knowledge formalization tool to the (on-line) fault diagnosis tool. They have been successfully tested on engine dyno test benches at Siemens VDO Automotive. Thus, the measurements made during car ECU (Electronic Control Unit) calibration are known to be valid (leading to a better quality for the calibration process) or the fault that occurred is clearly identified (reducing time to get back to normal behavior).